Prompt Injection on CSQA (ASR, MDSR)
62ASRInspector
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InspectorArchitecture=Tree2026.05 | 62 | 40 | |
| AgentXposedArchitecture=Star2026.05 | 61.3 | 38.3 | |
| AgentSafeArchitecture=Star2026.05 | 60 | 38.3 | |
| AgentXposedArchitecture=Random2026.05 | 59.7 | 41.7 | |
| No DefenseArchitecture=Star2026.05 | 56.3 | 43.3 | |
| ChallengerArchitecture=Star2026.05 | 56 | 45 | |
| AgentSafeArchitecture=Chain2026.05 | 52.3 | 51.7 | |
| No DefenseArchitecture=Chain2026.05 | 52 | 46.7 | |
| InspectorArchitecture=Chain2026.05 | 50.7 | 51.7 | |
| No DefenseArchitecture=Tree2026.05 | 50 | 56.7 | |
| ChallengerArchitecture=Random2026.05 | 49.3 | 50 | |
| AgentXposedArchitecture=Tree2026.05 | 49 | 53.3 | |
| AgentSafeArchitecture=Tree2026.05 | 48 | 56.7 | |
| ChallengerArchitecture=Tree2026.05 | 46 | 56.7 | |
| AgentSafeArchitecture=Random2026.05 | 45.3 | 55 | |
| No DefenseArchitecture=Random2026.05 | 44.7 | 55 | |
| ChallengerArchitecture=Chain2026.05 | 44.4 | 60 | |
| InspectorArchitecture=Star2026.05 | 43.7 | 56.7 | |
| AgentXposedArchitecture=Chain2026.05 | 43.3 | 55 | |
| InspectorArchitecture=Random2026.05 | 36.3 | 63.3 | |
| GUARDIANArchitecture=Chain2026.05 | 31.3 | 70 | |
| GUARDIANArchitecture=Random2026.05 | 31 | 71.7 | |
| INFA-GuardArchitecture=Chain2026.05 | 29.7 | 71.7 | |
| GUARDIANArchitecture=Tree2026.05 | 29.3 | 70 | |
| INFA-GuardArchitecture=Random2026.05 | 28.7 | 71.7 | |
| GUARDIANArchitecture=Star2026.05 | 27.7 | 73.3 | |
| G-SafeguardArchitecture=Tree2026.05 | 27.3 | 73.3 | |
| G-SafeguardArchitecture=Random2026.05 | 27.3 | 75 | |
| G-SafeguardArchitecture=Chain2026.05 | 26.7 | 75 | |
| INFA-GuardArchitecture=Tree2026.05 | 25.3 | 71.7 | |
| G-SafeguardArchitecture=Star2026.05 | 24.7 | 75 | |
| INFA-GuardArchitecture=Star2026.05 | 24.7 | 75 | |
| PropGuardArchitecture=Star2026.05 | 21 | 80 | |
| PropGuardArchitecture=Random2026.05 | 18.3 | 83.3 | |
| PropGuardArchitecture=Tree2026.05 | 18 | 83.3 | |
| PropGuardArchitecture=Chain2026.05 | 16 | 75 |